QSVM vs Classical SVM for Classification BenchmarksQuantum Kernels Against Classical Baselines
Audit a 569-sample QSVM benchmark against classical SVM.
Compare Quantum Kernels with a Classical SVM Baseline
Comparison Setup
A classical SVM separates classes by maximizing the margin in feature space. The source QSVM use case uses a binary breast-cancer classification benchmark with 569 samples and 30 FNA-derived features, scaled to the quantum feature range.
QSVM keeps the SVM pipeline but uses a quantum kernel: inputs are encoded into quantum states and the kernel is estimated from state overlaps. In the PoC, the quantum-kernel run uses 4 qubits and 2 layers, then reports accuracy, balanced accuracy, Macro F1, and ROC-AUC against classical baselines.
- • Does the QSVM run reach the reported 96% accuracy while preserving class balance?
- • Which preprocessing, feature map, 4-qubit encoding, and 2-layer circuit were used?
- • Are results reproducible with fixed seeds and documented simulator assumptions?
Why compare them
- Classical SVM gives a familiar baseline with well-understood margin and kernel behavior.
- QSVM replaces the kernel evaluation with quantum-state overlaps to test whether the embedding helps the task.
- A shared benchmark prevents quantum-kernel claims from being evaluated without a classical reference.
Limits to keep explicit
- Quantum-kernel improvements are not guaranteed and depend on the dataset and feature map.
- Current devices add gate and measurement noise that can degrade classification quality.
- Simulator runs avoid hardware noise but add their own computational cost and scaling limits.
Comparison Outputs
The source PoC reports 96% accuracy, 95% balanced accuracy, 0.95 Macro F1, and 0.99 ROC-AUC on the simulator run.
Dataset
569 samples
Features
30 FNA features
Quantum kernel
4 qubits, 2 layers
Accuracy
96%
Who should use this comparison
Use it when a quantum-kernel result needs a clear classical SVM baseline before it is trusted.
569
Breast-cancer samples
30
FNA-derived features
4 qubits
Quantum-kernel width
0.99
Reported ROC-AUC
ML benchmark teams
For teams that need the same data split, preprocessing, and metrics across quantum and classical kernels.
Applied quantum researchers
For researchers testing whether a quantum feature map changes classification quality.
Product and R&D leads
For teams deciding whether a QSVM result is strong enough to justify further experiments.
How it works
One dataset, two kernel pipelines, one reproducible comparison.
Prepare Data
Use the same split, scaling, labels, and 30-feature FNA dataset for both SVM variants
Run SVM
Train the classical SVM baseline and record the core classification metrics
Run QSVM
Encode features into a 4-qubit, 2-layer quantum kernel and evaluate state overlaps
Compare
Review 96% accuracy, 95% balanced accuracy, 0.95 Macro F1, 0.99 ROC-AUC, and assumptions
Export
Download the methods, plots, code, and reproducible benchmark report
Benchmark QSVM Against Classical SVM
Run the 569-sample QSVM benchmark and review 96% accuracy, 95% balanced accuracy, 0.95 Macro F1, and 0.99 ROC-AUC.
Try Your First Use Case for Free